Hadoop2.7.6_07_HA高可用

1. Hadoop的HA机制

　　前言：正式引入HA机制是从hadoop2.0开始，之前的版本中没有HA机制

1.1. HA的运作机制

（1）hadoop-HA集群运作机制介绍

　　所谓HA，即高可用（7*24小时不中断服务）

　　实现高可用最关键的是消除单点故障

　　hadoop-ha严格来说应该分成各个组件的HA机制——HDFS的HA、YARN的HA

（2）HDFS的HA机制详解

　　通过双namenode消除单点故障

　　双namenode协调工作的要点：

　　A、元数据管理方式需要改变：

　　　　内存中各自保存一份元数据

　　　　 Edits日志只能有一份，只有Active状态的namenode节点可以做写操作

　　　　两个namenode都可以读取edits

　　　　共享的edits放在一个共享存储中管理（qjournal和NFS两个主流实现）

　　B、需要一个状态管理功能模块

　　　　实现了一个zkfailover，常驻在每一个namenode所在的节点

　　　　每一个zkfailover负责监控自己所在namenode节点，利用zk进行状态标识

　　　　当需要进行状态切换时，由zkfailover来负责切换

　　　　 切换时需要防止brain split现象的发生

1.2. HDFS-HA图解

2. 主机规划

主机名称	外网IP	内网IP	操作系统	备注	安装软件	运行进程
mini01	10.0.0.111	172.16.1.111	CentOS 7.4	ssh port:22	jdk、hadoop	NameNode、DFSZKFailoverController(zkfc)
mini02	10.0.0.112	172.16.1.112	CentOS 7.4	ssh port:22	jdk、hadoop	NameNode、DFSZKFailoverController(zkfc)
mini03	10.0.0.113	172.16.1.113	CentOS 7.4	ssh port:22	jdk、hadoop、zookeeper	ResourceManager
mini04	10.0.0.114	172.16.1.114	CentOS 7.4	ssh port:22	jdk、hadoop、zookeeper	ResourceManager
mini05	10.0.0.115	172.16.1.115	CentOS 7.4	ssh port:22	jdk、hadoop、zookeeper	DataNode、NodeManager、JournalNode、QuorumPeerMain
mini06	10.0.0.116	172.16.1.116	CentOS 7.4	ssh port:22	jdk、hadoop、zookeeper	DataNode、NodeManager、JournalNode、QuorumPeerMain
mini07	10.0.0.117	172.16.1.117	CentOS 7.4	ssh port:22	jdk、hadoop、zookeeper	DataNode、NodeManager、JournalNode、QuorumPeerMain

注意：针对HA模式，就不需要SecondaryNameNode了，因为STANDBY状态的namenode会负责做checkpoint

　　

Linux添加hosts信息，保证每台都可以相互ping通

 [root@mini01 ~]# cat /etc/hosts
 127.0.0.1   localhost localhost.localdomain localhost4 localhost4.localdomain4
 ::         localhost localhost.localdomain localhost6 localhost6.localdomain6

 10.0.0.111    mini01
 10.0.0.112    mini02
 10.0.0.113    mini03
 10.0.0.114    mini04
 10.0.0.115    mini05
 10.0.0.116    mini06
 10.0.0.117    mini07

Windows的hosts文件修改

 # 文件位置C:\Windows\System32\drivers\etc   在hosts中追加如下内容
 …………………………………………
 10.0.0.111    mini01
 10.0.0.112    mini02
 10.0.0.113    mini03
 10.0.0.114    mini04
 10.0.0.115    mini05
 10.0.0.116    mini06
 10.0.0.117    mini07

3. 添加用户账号

 # 使用一个专门的用户，避免直接使用root用户
 # 添加用户、指定家目录并指定用户密码
 useradd -d /app yun && echo '' | /usr/bin/passwd --stdin yun
 # sudo提权
 echo "yun  ALL=(ALL)       NOPASSWD: ALL" >>  /etc/sudoers
 # 让其它普通用户可以进入该目录查看信息
 chmod  /app/

4. 实现yun用户免秘钥登录

 要求：根据规划实现 mini01 到 mini01、mini02、mini03、mini04、mini05、mini06、mini07 免秘钥登录
               实现 mini02 到 mini01、mini02、mini03、mini04、mini05、mini06、mini07 免秘钥登录
               实现 mini03 到 mini01、mini02、mini03、mini04、mini05、mini06、mini07 免秘钥登录
               实现 mini04 到 mini01、mini02、mini03、mini04、mini05、mini06、mini07 免秘钥登录
               实现 mini05 到 mini01、mini02、mini03、mini04、mini05、mini06、mini07 免秘钥登录
               实现 mini06 到 mini01、mini02、mini03、mini04、mini05、mini06、mini07 免秘钥登录
               实现 mini07 到 mini01、mini02、mini03、mini04、mini05、mini06、mini07 免秘钥登录

 # 可以使用ip也可以是hostname  但是由于我们计划使用的是 hostname 方式交互，所以使用hostname
 # 同时hostname方式分发，可以通过hostname远程登录，也可以IP远程登录

具体过程就不多说了，请参见 Hadoop2.7.6_01_部署

5. Jdk【java8】

具体过程就不多说了，请参见 Hadoop2.7.6_01_部署

6. Zookeeper部署

根据规划zookeeper部署在mini03、mini04、mini05、mini06、mini07上

6.1. 配置信息

 [yun@mini03 conf]$ pwd
 /app/zookeeper/conf
 [yun@mini03 conf]$ vim zoo.cfg
 #单个客户端与单台服务器之间的连接数的限制，是ip级别的，默认是60，如果设置为0，那么表明不作任何限制。
 maxClientCnxns=
 # The number of milliseconds of each tick
 tickTime=
 # The number of ticks that the initial
 # synchronization phase can take
 initLimit=
 # The number of ticks that can pass between
 # sending a request and getting an acknowledgement
 syncLimit=
 # the directory where the snapshot is stored.
 # do not use /tmp for storage, /tmp here is just
 # example sakes.
 # dataDir=/tmp/zookeeper
 dataDir=/app/bigdata/zookeeper/data
 # the port at which the clients will connect
 clientPort=
 #
 # Be sure to read the maintenance section of the
 # administrator guide before turning on autopurge.
 #
 # http://zookeeper.apache.org/doc/current/zookeeperAdmin.html#sc_maintenance
 #
 # The number of snapshots to retain in dataDir
 #autopurge.snapRetainCount=
 # Purge task interval in hours
 # Set to "" to disable auto purge feature
 #autopurge.purgeInterval=

 # leader和follow通信端口和投票选举端口
 server.=mini03::
 server.=mini04::
 server.=mini05::
 server.=mini06::
 server.=mini07::

6.2. 添加myid文件

 [yun@mini03 data]$ pwd
 /app/bigdata/zookeeper/data
 [yun@mini03 data]$ vim myid   # 其中mini03的myid 为3；mini04的myid 为4；mini05的myid 为5；mini06的myid 为6；mini07的myid 为7
 

6.3. 启动zk服务

 # 依次在启动mini03、mini04、mini05、mini06、mini07 zk服务
 [yun@mini03 ~]$ cd zookeeper/bin/
 [yun@mini03 bin]$ pwd
 /app/zookeeper/bin
 [yun@mini03 bin]$ ll
 total
 -rwxr-xr-x  yun yun    Oct     README.txt
 -rwxr-xr-x  yun yun   Oct     zkCleanup.sh
 -rwxr-xr-x  yun yun   Oct     zkCli.cmd
 -rwxr-xr-x  yun yun   Oct     zkCli.sh
 -rwxr-xr-x  yun yun   Oct     zkEnv.cmd
 -rwxr-xr-x  yun yun   Oct     zkEnv.sh
 -rwxr-xr-x  yun yun   Oct     zkServer.cmd
 -rwxr-xr-x  yun yun   Oct     zkServer.sh
 -rw-rw-r--  yun yun  Jun  : zookeeper.out
 [yun@mini03 bin]$ ./zkServer.sh start
 JMX enabled by default
 Using config: /app/zookeeper/bin/../conf/zoo.cfg
 Starting zookeeper ... STARTED

6.4. 查询运行状态

 # 其中mini03、mini04、mini06、mini07状态如下
 [yun@mini03 bin]$ ./zkServer.sh status
 JMX enabled by default
 Using config: /app/zookeeper/bin/../conf/zoo.cfg
 Mode: follower

 # 其中mini05 状态如下
 [yun@mini05 bin]$ ./zkServer.sh status
 JMX enabled by default
 Using config: /app/zookeeper/bin/../conf/zoo.cfg
 Mode: leader

PS：4个follower 1个leader

7. Hadoop部署与配置修改

　　注意：每台机器的Hadoop以及配置相同

7.1. 部署

 [yun@mini01 software]$ pwd
 /app/software
 [yun@mini01 software]$ ll
 total
 -rw-r--r--  yun yun  Jun   : CentOS-.4_hadoop-2.7..tar.gz
 [yun@mini01 software]$ tar xf CentOS-.4_hadoop-2.7..tar.gz
 [yun@mini01 software]$ mv hadoop-2.7./ /app/
 [yun@mini01 software]$ cd
 [yun@mini01 ~]$ ln -s hadoop-2.7./ hadoop
 [yun@mini01 ~]$ ll
 total
 lrwxrwxrwx   yun yun    Jun   : hadoop -> hadoop-2.7./
 drwxr-xr-x   yun yun   Jun   : hadoop-2.7.
 lrwxrwxrwx   yun yun    May  : jdk -> jdk1..0_112
 drwxr-xr-x   yun yun   Sep    jdk1..0_112

7.2. 环境变量

 [root@mini01 profile.d]# pwd
 /etc/profile.d
 [root@mini01 profile.d]# vim hadoop.sh
 export HADOOP_HOME="/app/hadoop"
 export PATH=$HADOOP_HOME/bin:$HADOOP_HOME/sbin:$PATH

 [root@mini01 profile.d]# source /etc/profile  # 生效

7.3. core-site.xml

 [yun@mini01 hadoop]$ pwd
 /app/hadoop/etc/hadoop
 [yun@mini01 hadoop]$ vim core-site.xml
 <?xml version="1.0" encoding="UTF-8"?>
 <?xml-stylesheet type="text/xsl" href="configuration.xsl"?>
 <!--
 ……………………
 -->

 <!-- Put site-specific property overrides in this file. -->

 <configuration>
   <!-- 指定hdfs的nameservice为bi -->
   <property>
     <name>fs.defaultFS</name>
     <value>hdfs://bi/</value>
   </property>

   <!-- 指定hadoop临时目录 -->
   <property>
     <name>hadoop.tmp.dir</name>
     <value>/app/hadoop/tmp</value>
   </property>

   <!-- 指定zookeeper地址 -->
   <property>
     <name>ha.zookeeper.quorum</name>
     <value>mini03:,mini04:,mini05:,mini06:,mini07:</value>
   </property>

 </configuration>

7.4. hdfs-site.xml

 [yun@mini01 hadoop]$ pwd
 /app/hadoop/etc/hadoop
 [yun@mini01 hadoop]$ vim hdfs-site.xml
 <?xml version="1.0" encoding="UTF-8"?>
 <?xml-stylesheet type="text/xsl" href="configuration.xsl"?>
 <!--
   ……………………
 -->

 <!-- Put site-specific property overrides in this file. -->

 <configuration>
   <!--指定hdfs的nameservice为bi，需要和core-site.xml中的保持一致 -->
   <property>
     <name>dfs.nameservices</name>
     <value>bi</value>
   </property>

   <!-- bi下面有两个NameNode，分别是nn1，nn2 -->
   <property>
     <name>dfs.ha.namenodes.bi</name>
     <value>nn1,nn2</value>
   </property>

   <!-- nn1的RPC通信地址 -->
   <property>
     <name>dfs.namenode.rpc-address.bi.nn1</name>
     <value>mini01:</value>
   </property>
   <!-- nn1的http通信地址 -->
   <property>
     <name>dfs.namenode.http-address.bi.nn1</name>
     <value>mini01:</value>
   </property>

   <!-- nn2的RPC通信地址 -->
   <property>
     <name>dfs.namenode.rpc-address.bi.nn2</name>
     <value>mini02:</value>
   </property>
   <!-- nn2的http通信地址 -->
   <property>
     <name>dfs.namenode.http-address.bi.nn2</name>
     <value>mini02:</value>
   </property>

   <!-- 指定NameNode的edits元数据在JournalNode上的存放位置 -->
   <property>
     <name>dfs.namenode.shared.edits.dir</name>
     <value>qjournal://mini05:8485;mini06:8485;mini07:8485/bi</value>
   </property>

   <!-- 指定JournalNode在本地磁盘存放数据的位置 -->
   <property>
     <name>dfs.journalnode.edits.dir</name>
     <value>/app/hadoop/journaldata</value>
   </property>

   <!-- 开启NameNode失败自动切换 -->
   <property>
     <name>dfs.ha.automatic-failover.enabled</name>
     <value>true</value>
   </property>

   <!-- 配置失败自动切换实现方式 -->
   <property>
     <name>dfs.client.failover.proxy.provider.bi</name>
     <value>org.apache.hadoop.hdfs.server.namenode.ha.ConfiguredFailoverProxyProvider</value>
   </property>

   <!-- 配置隔离机制方法，多个机制用换行分割，即每个机制暂用一行-->
   <!-- 其中shell(/bin/true) 表示可执行一个脚本  比如 shell(/app/yunwei/hadoop_fence.sh) -->
   <property>
     <name>dfs.ha.fencing.methods</name>
     <value>
       sshfence
       shell(/bin/true)
     </value>
   </property>

   <!-- 使用sshfence隔离机制时需要ssh免登陆 -->
   <property>
     <name>dfs.ha.fencing.ssh.private-key-files</name>
     <value>/app/.ssh/id_rsa</value>
   </property>

   <!-- 配置sshfence隔离机制超时时间 单位：毫秒 -->
   <property>
     <name>dfs.ha.fencing.ssh.connect-timeout</name>
     <value></value>
   </property>

 </configuration>

7.5. mapred-site.xml

 [yun@mini01 hadoop]$ pwd
 /app/hadoop/etc/hadoop
 [yun@mini01 hadoop]$ cp -a mapred-site.xml.template mapred-site.xml
 [yun@mini01 hadoop]$ vim mapred-site.xml
 <?xml version="1.0"?>
 <?xml-stylesheet type="text/xsl" href="configuration.xsl"?>
 <!--
   ……………………
 -->

 <!-- Put site-specific property overrides in this file. -->

 <configuration>
   <!-- 指定mr框架为yarn方式 -->
   <property>
     <name>mapreduce.framework.name</name>
     <value>yarn</value>
   </property>

 </configuration>

7.6. yarn-site.xml

 [yun@mini01 hadoop]$ pwd
 /app/hadoop/etc/hadoop
 [yun@mini01 hadoop]$ vim yarn-site.xml
 <?xml version="1.0"?>
 <!--
   ……………………
 -->
 <configuration>

 <!-- Site specific YARN configuration properties -->
   <!-- 开启RM高可用 -->
   <property>
     <name>yarn.resourcemanager.ha.enabled</name>
     <value>true</value>
   </property>

   <!-- 指定RM的cluster id -->
   <property>
     <name>yarn.resourcemanager.cluster-id</name>
     <value>yrc</value>
   </property>

   <!-- 指定RM的名字 -->
   <property>
     <name>yarn.resourcemanager.ha.rm-ids</name>
     <value>rm1,rm2</value>
   </property>

   <!-- 分别指定RM的地址 -->
   <property>
     <name>yarn.resourcemanager.hostname.rm1</name>
     <value>mini03</value>
   </property>
   <property>
     <name>yarn.resourcemanager.hostname.rm2</name>
     <value>mini04</value>
   </property>

   <!-- 指定zk集群地址 -->
   <property>
     <name>yarn.resourcemanager.zk-address</name>
     <value>mini03:,mini04:,mini05:,mini06:,mini07:</value>
   </property>

   <!-- reduce 获取数据的方式 -->
   <property>
     <name>yarn.nodemanager.aux-services</name>
     <value>mapreduce_shuffle</value>
   </property>

 </configuration>

7.7. 修改slaves

slaves是指定子节点的位置，因为要在mini01上启动HDFS、在mini03启动yarn，所以mini01上的slaves文件指定的是datanode的位置，mini03上的slaves文件指定的是nodemanager的位置

 [yun@mini01 hadoop]$ pwd
 /app/hadoop/etc/hadoop
 [yun@mini01 hadoop]$ vim slaves
 mini05
 mini06
 mini07

PS：改后配置后，将这些配置拷到其他Hadoop机器

8. 启动相关服务

　　注意：第一次启动时严格按照下面的步骤！！！！！！！

8.1. 启动zookeeper集群

前面已经启动了，这里就不说了

8.2. 启动journalnode

 # 根据规划在mini05、mini06、mini07 启动
 # 在第一次格式化的时候需要先启动journalnode   之后就不必了
 [yun@mini05 ~]$ hadoop-daemon.sh start journalnode  # 已经配置环境变量，所以不用进入到响应的目录
 starting journalnode, logging to /app/hadoop-2.7./logs/hadoop-yun-journalnode-mini05.out
 [yun@mini05 ~]$ jps
  QuorumPeerMain
  Jps
  JournalNode

8.3. 格式化HDFS

 # 在mini01上执行命令
 [yun@mini01 ~]$ hdfs namenode -format
 // :: INFO namenode.NameNode: STARTUP_MSG:
 /************************************************************
 STARTUP_MSG: Starting NameNode
 STARTUP_MSG:   host = mini01/10.0.0.111
 STARTUP_MSG:   args = [-format]
 STARTUP_MSG:   version = 2.7.6
 STARTUP_MSG:   classpath = ………………
 STARTUP_MSG:   build = Unknown -r Unknown; compiled by 'root' on 2018-06-08T08:30Z
 STARTUP_MSG:   java = 1.8.0_112
 ************************************************************/
 // :: INFO namenode.NameNode: registered UNIX signal handlers for [TERM, HUP, INT]
 // :: INFO namenode.NameNode: createNameNode [-format]
 Formatting using clusterid: CID-2385f26e-72e6--aa09-47848b5ba4be
 // :: INFO namenode.FSNamesystem: No KeyProvider found.
 // :: INFO namenode.FSNamesystem: fsLock is fair: true
 // :: INFO namenode.FSNamesystem: Detailed lock hold time metrics enabled: false
 // :: INFO blockmanagement.DatanodeManager: dfs.block.invalidate.limit=
 // :: INFO blockmanagement.DatanodeManager: dfs.namenode.datanode.registration.ip-hostname-check=true
 // :: INFO blockmanagement.BlockManager: dfs.namenode.startup.delay.block.deletion.sec is set to :::00.000
 // :: INFO blockmanagement.BlockManager: The block deletion will start around  Jun  ::
 // :: INFO util.GSet: Computing capacity for map BlocksMap
 // :: INFO util.GSet: VM type       = -bit
 // :: INFO util.GSet: 2.0% max memory 966.7 MB = 19.3 MB
 // :: INFO util.GSet: capacity      = ^ =  entries
 // :: INFO blockmanagement.BlockManager: dfs.block.access.token.enable=false
 // :: INFO blockmanagement.BlockManager: defaultReplication         =
 // :: INFO blockmanagement.BlockManager: maxReplication             =
 // :: INFO blockmanagement.BlockManager: minReplication             =
 // :: INFO blockmanagement.BlockManager: maxReplicationStreams      =
 // :: INFO blockmanagement.BlockManager: replicationRecheckInterval =
 // :: INFO blockmanagement.BlockManager: encryptDataTransfer        = false
 // :: INFO blockmanagement.BlockManager: maxNumBlocksToLog          =
 // :: INFO namenode.FSNamesystem: fsOwner             = yun (auth:SIMPLE)
 // :: INFO namenode.FSNamesystem: supergroup          = supergroup
 // :: INFO namenode.FSNamesystem: isPermissionEnabled = true
 // :: INFO namenode.FSNamesystem: Determined nameservice ID: bi
 // :: INFO namenode.FSNamesystem: HA Enabled: true
 // :: INFO namenode.FSNamesystem: Append Enabled: true
 // :: INFO util.GSet: Computing capacity for map INodeMap
 // :: INFO util.GSet: VM type       = -bit
 // :: INFO util.GSet: 1.0% max memory 966.7 MB = 9.7 MB
 // :: INFO util.GSet: capacity      = ^ =  entries
 // :: INFO namenode.FSDirectory: ACLs enabled? false
 // :: INFO namenode.FSDirectory: XAttrs enabled? true
 // :: INFO namenode.FSDirectory: Maximum size of an xattr:
 // :: INFO namenode.NameNode: Caching file names occuring more than  times
 // :: INFO util.GSet: Computing capacity for map cachedBlocks
 // :: INFO util.GSet: VM type       = -bit
 // :: INFO util.GSet: 0.25% max memory 966.7 MB = 2.4 MB
 // :: INFO util.GSet: capacity      = ^ =  entries
 // :: INFO namenode.FSNamesystem: dfs.namenode.safemode.threshold-pct = 0.9990000128746033
 // :: INFO namenode.FSNamesystem: dfs.namenode.safemode.min.datanodes =
 // :: INFO namenode.FSNamesystem: dfs.namenode.safemode.extension     =
 // :: INFO metrics.TopMetrics: NNTop conf: dfs.namenode.top.window.num.buckets =
 // :: INFO metrics.TopMetrics: NNTop conf: dfs.namenode.top.num.users =
 // :: INFO metrics.TopMetrics: NNTop conf: dfs.namenode.top.windows.minutes = ,,
 // :: INFO namenode.FSNamesystem: Retry cache on namenode is enabled
 // :: INFO namenode.FSNamesystem: Retry cache will use 0.03 of total heap and retry cache entry expiry time is  millis
 // :: INFO util.GSet: Computing capacity for map NameNodeRetryCache
 // :: INFO util.GSet: VM type       = -bit
 // :: INFO util.GSet: 0.029999999329447746% max memory 966.7 MB = 297.0 KB
 // :: INFO util.GSet: capacity      = ^ =  entries
 // :: INFO namenode.FSImage: Allocated new BlockPoolId: BP--10.0.0.111-
 // :: INFO common.Storage: Storage directory /app/hadoop/tmp/dfs/name has been successfully formatted.
 // :: INFO namenode.FSImageFormatProtobuf: Saving image file /app/hadoop/tmp/dfs/name/current/fsimage.ckpt_0000000000000000000 using no compression
 // :: INFO namenode.FSImageFormatProtobuf: Image file /app/hadoop/tmp/dfs/name/current/fsimage.ckpt_0000000000000000000 of size  bytes saved in  seconds.
 // :: INFO namenode.NNStorageRetentionManager: Going to retain  images with txid >=
 // :: INFO util.ExitUtil: Exiting with status
 // :: INFO namenode.NameNode: SHUTDOWN_MSG:
 /************************************************************
 SHUTDOWN_MSG: Shutting down NameNode at mini01/10.0.0.111
 ************************************************************/

拷贝到mini02

 #格式化后会在根据core-site.xml中的hadoop.tmp.dir配置生成个文件，这里我配置的是/app/hadoop/tmp，然后将/app/hadoop/tmp拷贝到mini02的/app/hadoop/下。
 # 方法1：
 [yun@mini01 hadoop]$ pwd
 /app/hadoop
 [yun@mini01 hadoop]$ scp -r tmp/ yun@mini02:/app/hadoop
 VERSION                           %     .4KB/s   :
 seen_txid                         %         .0KB/s   :
 fsimage_0000000000000000000.md5   %       .7KB/s   :
 fsimage_0000000000000000000       %     .1KB/s   : 

 ##########################
 # 方法2：##也可以这样，建议hdfs namenode -bootstrapStandby  # 不过需要mini02的Hadoop起来才行

8.4. 格式化ZKFC

 #在mini01上执行一次即可
 [yun@mini01 ~]$ hdfs zkfc -formatZK
 // :: INFO tools.DFSZKFailoverController: Failover controller configured for NameNode NameNode at mini01/10.0.0.111:
 // :: INFO zookeeper.ZooKeeper: Client environment:zookeeper.version=3.4.-, built on // : GMT
 // :: INFO zookeeper.ZooKeeper: Client environment:host.name=mini01
 // :: INFO zookeeper.ZooKeeper: Client environment:java.version=1.8.0_112
 // :: INFO zookeeper.ZooKeeper: Client environment:java.vendor=Oracle Corporation
 // :: INFO zookeeper.ZooKeeper: Client environment:java.home=/app/jdk1..0_112/jre
 // :: INFO zookeeper.ZooKeeper: Client environment:java.class.path=……………………
 // :: INFO zookeeper.ZooKeeper: Client environment:java.library.path=/app/hadoop-2.7./lib/native
 // :: INFO zookeeper.ZooKeeper: Client environment:java.io.tmpdir=/tmp
 // :: INFO zookeeper.ZooKeeper: Client environment:java.compiler=<NA>
 // :: INFO zookeeper.ZooKeeper: Client environment:os.name=Linux
 // :: INFO zookeeper.ZooKeeper: Client environment:os.arch=amd64
 // :: INFO zookeeper.ZooKeeper: Client environment:os.version=3.10.-.el7.x86_64
 // :: INFO zookeeper.ZooKeeper: Client environment:user.name=yun
 // :: INFO zookeeper.ZooKeeper: Client environment:user.home=/app
 // :: INFO zookeeper.ZooKeeper: Client environment:user.dir=/app/hadoop-2.7.
 // :: INFO zookeeper.ZooKeeper: Initiating client connection, connectString=mini03:,mini04:,mini05:,mini06:,mini07: sessionTimeout= watcher=org.apache.hadoop.ha.ActiveStandbyElector$WatcherWithClientRef@7f3b84b8
 // :: INFO zookeeper.ClientCnxn: Opening socket connection to server mini04/10.0.0.114:. Will not attempt to authenticate using SASL (unknown error)
 // :: INFO zookeeper.ClientCnxn: Socket connection established to mini04/10.0.0.114:, initiating session
 // :: INFO zookeeper.ClientCnxn: Session establishment complete on server mini04/10.0.0.114:, sessionid = 0x4644fff9cb80000, negotiated timeout =
 // :: INFO ha.ActiveStandbyElector: Session connected.
 // :: INFO ha.ActiveStandbyElector: Successfully created /hadoop-ha/bi in ZK.
 // :: INFO zookeeper.ZooKeeper: Session: 0x4644fff9cb80000 closed
 // :: INFO zookeeper.ClientCnxn: EventThread shut down

8.5. 启动HDFS

 # 在mini01上执行
 [yun@mini01 ~]$ start-dfs.sh
 Starting namenodes on [mini01 mini02]
 mini01: starting namenode, logging to /app/hadoop-2.7./logs/hadoop-yun-namenode-mini01.out
 mini02: starting namenode, logging to /app/hadoop-2.7./logs/hadoop-yun-namenode-mini02.out
 mini07: starting datanode, logging to /app/hadoop-2.7./logs/hadoop-yun-datanode-mini07.out
 mini06: starting datanode, logging to /app/hadoop-2.7./logs/hadoop-yun-datanode-mini06.out
 mini05: starting datanode, logging to /app/hadoop-2.7./logs/hadoop-yun-datanode-mini05.out
 Starting journal nodes [mini05 mini06 mini07]
 mini07: journalnode running as process . Stop it first.
 mini06: journalnode running as process . Stop it first.
 mini05: journalnode running as process . Stop it first.
 Starting ZK Failover Controllers on NN hosts [mini01 mini02]
 mini01: starting zkfc, logging to /app/hadoop-2.7./logs/hadoop-yun-zkfc-mini01.out
 mini02: starting zkfc, logging to /app/hadoop-2.7./logs/hadoop-yun-zkfc-mini02.out

8.6. 启动YARN

 #####注意#####：是在mini03上执行start-yarn.sh，把namenode和resourcemanager分开是因为性能问题
 # 因为他们都要占用大量资源，所以把他们分开了，他们分开了就要分别在不同的机器上启动
 [yun@mini03 ~]$ start-yarn.sh
 starting yarn daemons
 starting resourcemanager, logging to /app/hadoop-2.7./logs/yarn-yun-resourcemanager-mini03.out
 mini06: starting nodemanager, logging to /app/hadoop-2.7./logs/yarn-yun-nodemanager-mini06.out
 mini07: starting nodemanager, logging to /app/hadoop-2.7./logs/yarn-yun-nodemanager-mini07.out
 mini05: starting nodemanager, logging to /app/hadoop-2.7./logs/yarn-yun-nodemanager-mini05.out

 ################################
 # 在mini04启动 resourcemanager
 [yun@mini04 ~]$ yarn-daemon.sh start resourcemanager  # 也可用start-yarn.sh
 starting resourcemanager, logging to /app/hadoop-2.7./logs/yarn-yun-resourcemanager-mini04.out

8.7. 启动说明

 # 第一次启动的时候请严格按照上面的步骤【第一次涉及格式化问题】
 # 第二次以及之后，步骤为： 启动zookeeper、HDFS、YARN

9. 浏览访问

9.1. Hdfs访问

9.1.1. 正常情况访问

 http://mini01:50070    

 http://mini02:50070    

9.1.2. mini01挂了Active自动切换

 # mini01操作
 [yun@mini01 ~]$ jps
  DFSZKFailoverController
  NameNode
  Jps
 [yun@mini01 ~]$ kill
 [yun@mini01 ~]$ jps
  DFSZKFailoverController
  Jps

Namenode挂了所以mini01不能访问

 http://mini02:50070    

可见Hadoop已经切换过去了，之后mini01即使起来了，状态也只能为standby 。

9.2. Yarn访问

 http://mini03:8088    

 http://mini04:8088
 会直接跳转到http://mini03:8088/    

# 该图从其他地方截取，所以不怎么匹配

 # Linux下访问
 [yun@mini01 ~]$ curl mini04:
 This is standby RM. The redirect url is: http://mini03:8088/

HA完毕

10. 集群运维测试

10.1. Haadmin与状态切换管理

 [yun@mini01 ~]$ hdfs haadmin
 Usage: haadmin
     [-transitionToActive [--forceactive] <serviceId>]
     [-transitionToStandby <serviceId>]
     [-failover [--forcefence] [--forceactive] <serviceId> <serviceId>]
     [-getServiceState <serviceId>]
     [-checkHealth <serviceId>]
     [-help <command>]

 Generic options supported are
 -conf <configuration file>     specify an application configuration file
 -D <property=value>            use value for given property
 -fs <local|namenode:port>      specify a namenode
 -jt <local|resourcemanager:port>    specify a ResourceManager
 -files <comma separated list of files>    specify comma separated files to be copied to the map reduce cluster
 -libjars <comma separated list of jars>    specify comma separated jar files to include in the classpath.
 -archives <comma separated list of archives>    specify comma separated archives to be unarchived on the compute machines.

 The general command line syntax is
 bin/hadoop command [genericOptions] [commandOptions]

可以看到，状态操作的命令示例：

 # 查看namenode工作状态
 hdfs haadmin -getServiceState nn1

 # 将standby状态namenode切换到active
 hdfs haadmin -transitionToActive nn1

 # 将active状态namenode切换到standby
 hdfs haadmin -transitionToStandby nn2

10.2. 测试集群工作状态的一些指令

 测试集群工作状态的一些指令 ：
 hdfs dfsadmin -report     查看hdfs的各节点状态信息
 hdfs haadmin -getServiceState nn1   # hdfs haadmin -getServiceState nn2      获取一个namenode节点的HA状态
 hadoop-daemon.sh start namenode  单独启动一个namenode进程
 hadoop-daemon.sh start zkfc   单独启动一个zkfc进程

10.3. Datanode动态上下线

Datanode动态上下线很简单，步骤如下：

　　a)   准备一台服务器，设置好环境

　　b)   部署hadoop的安装包，并同步集群配置

　　c)   联网上线，新datanode会自动加入集群

　　d)   如果是一次增加大批datanode，还应该做集群负载重均衡

10.4. 数据块的balance

启动balancer的命令：

 start-balancer.sh -threshold 

运行之后，会有Balancer进程出现：

上述命令设置了Threshold为8%，那么执行balancer命令的时候，首先统计所有DataNode的磁盘利用率的均值，然后判断如果某一个DataNode的磁盘利用率超过这个均值Threshold，那么将会把这个DataNode的block转移到磁盘利用率低的DataNode，这对于新节点的加入来说十分有用。Threshold的值为1到100之间，不显示的进行参数设置的话，默认是10。